Bike-Repositioning Using Volunteers: Crowd Sourcing with Choice Restriction

Jinjia Huang, Mabel C. Chou, Chung-Piaw TeoAAAI 2021 · DOI

Introduction

Bike-sharing systems (BSS), such as New York’s Citi Bike or Boston’s Blue Bikes, face a persistent spatial imbalance problem: bikes accumulate in business districts during the morning and vanish from residential areas, with the reverse occurring in the evening. Traditionally, operators use trucks for rebalancing, but this is expensive and environmentally unfriendly. While crowdsourced "Bike Angel" programs incentivize volunteers to move bikes, "fully flexible" systems often lead to redundant moves that do not significantly improve system health. This paper demonstrates that BSS efficiency can be improved by restricting volunteer choices to a sparse network of essential routes.

Methodology

The authors propose a two-stage approach combining offline strategic planning with online tactical deployment:

  1. Sparse Network Design (Offline): The BSS is modeled as a transshipment network. Using Distributionally Robust Optimization (DRO), the researchers identify a sparse set of high-priority arcs (routes). DRO is particularly effective here as it only requires the first two moments of demand data, making it robust against real-world uncertainty.
  2. State-Dependent Deployment (Online): A real-time algorithm monitors station inventory and offers volunteers a restricted set of destination stations connected via the pre-identified sparse network.

The study validated this approach using real-world data from the Boston Hubway system (now Blue Bikes), analyzing 60 stations during the morning commute period (7:00 AM – 1:00 PM).

Key Results

The results indicate that restricting volunteer choice "concentrates" the flow of bikes where they are most needed, avoiding the "noise" introduced by local, uncoordinated decisions in fully flexible systems.

Comparison of fulfilled rider demand and rebalance moves between fully flexible and sparse solutions across varying volunteer arrival rates (c) and outside options.

As shown in the figure above, the sparse network approach (red bars) drastically reduces the number of rebalancing moves required compared to a fully flexible system (blue bars). Remarkably, the system maintains nearly the same level of fulfilled rider demand despite these restrictions. The authors found that a network with only 480 arcs performed within 1% of a fully connected network with 3,540 arcs, proving that sparsity is sufficient for effective rebalancing.

Conclusion

By identifying a "backbone" of essential routes, BSS operators can guide volunteers to make more impactful moves without micromanagement. Future work could further refine these models by incorporating rider substitution behavior and more complex volunteer incentive structures, such as Multinomial Logit choice frameworks.

Summary updated

BibTeX
@inproceedings{huang2021bike,
  author = {Jinjia Huang and Mabel C. Chou and Chung-Piaw Teo},
  title = {{Bike-Repositioning Using Volunteers: Crowd Sourcing with Choice Restriction}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/aaai.v35i13.17407},
}
Figure